Meteorological Conditions Conducive to the Rapid Spread of the Deadly Wildfire in Eastern ,

K. Lagouvardos, V. Kotroni, T. M. Giannaros, and S. Dafis

xtreme fire behavior is a term extensively used in the literature Ewith various subjective defini- tions. Werth et al. (2016) came up with the following definition of ex- treme fire behavior: “fire spread other than steady surface spread, especially when it involves rapid increases”; they also stated that all three factors of the fire behavior triangle—fuels, weather, and topography—should be taken into account when framing perceptions of extreme fire behavior. Focusing on the weather conditions contributing to extreme fire behavior, Werth et al. (2016) in their review of previous studies identified four such elements: low relative humid- ity, strong surface wind, unstable air, and drought. Strong surface winds, high temperatures, and low relative Fig. 1. Map of the Attica region. The red asterisks denote the starting humidity develop during strong points of the two forest fires. Letters P, R, and N denote the locations of downslope winds (Whiteman 2000; the Penteli, , and Nea Makri automatic weather stations, respec- Huang et al. 2009; Mass and Ovens tively. Shading depicts the model-resolved topography (at 50-m inter- 2019). Such conditions occur particu- vals). A zoom over the area of interest is also provided as an inset map. larly often in the Mediterranean-type ecosystem of Southern California, where wildfires are clearly associated with strong 2018; Mass and Ovens 2019). In the western Mediter- downslope winds (Abatzoglou et al. 2013; Nauslar et al. ranean, large summer wildfires were also found to be

AFFILIATIONS: Lagouvardos, Kotroni, Giannaros, and Dafis— DOI:10.1175/BAMS-D-18-0231.1 National Observatory of , Athens, Greece A supplement to this article is available online (10.1175/BAMS-D-18-0231.2) CORRESPONDING AUTHOR: Konstantinos Lagouvardos, [email protected] ©2019 American Meteorological Society For information regarding reuse of this content and general The abstract for this article can be found in this issue, following the table copyright information, consult the AMS Copyright Policy. of contents.

PB | NOVEMBER 2019 AMERICAN METEOROLOGICAL SOCIETY NOVEMBER 2019 | 2137 Unauthenticated | Downloaded 10/05/21 05:16 PM UTC Fig. 2. Vehicles and houses burned within Mati area. (bottom-right) An aerial view of part of the coastline. The photos were taken 2 days after the disaster. (Source: Lekkas et al. 2018)

associated with strong continental dry winds (Ruffault burned and 305 burned vehicles, while the total area et al. 2017). In the eastern Mediterranean, and more burned reached 1,250 ha (Lekkas et al. 2018). A series precisely in Greece, Diakakis et al. (2016), who studied of photos depicting the catastrophic impact of the forest-fire-related fatalities for a 36-yr period, found wildland fire are shown in Fig. 2. that major events were related to high temperatures The rest of the paper is structured as follows. First (over 30°C), low relative humidity (less than 30%), and the prevailing meteorological conditions during the wind speeds exceeding 8 m s−1. event are briefly discussed, based on the combined The motivation of the present study is the recent analysis of automatic weather station (AWS) observa- deadly wildfire that took place in Attica, Greece, on tions, provided by the National Observatory of Athens 23 July 2018. At around 0900 UTC, a wildfire was (NOAAN; Lagouvardos et al. 2017), and model results ignited in a dense pine forest in the western part of from high-resolution simulations carried out with a Attica, spreading rapidly, but fortunately did not coupled fire–atmosphere modeling system. Then the result in any human casualties. Almost 5 h later, at results of the fire spread simulations are presented. 1355 UTC, a wildfire broke out in a mountainous Finally, the main findings are summarized and the forested area of the east of Attica (see Fig. 1 for the key points related to improving preparedness are location of the areas mentioned in the text). The rapid discussed, especially with respect to wildfires taking spread of the second wildfire toward the east affected place in the WUI. a wildland–urban interface (WUI) area and resulted in the death of 102 civilians in less than 3 h. The ex- BRIEF SYNOPTIC DESCRIPTION AND ceptionally high death toll establishes this event as the SURFACE OBSERVATIONS. The 0.25°-hori- second-deadliest weather-related disaster in Greece, zontal-resolution Global Data Assimilation System the major heat wave of July 1987 being the deadliest. (GDAS) tropospheric analyses data were employed for Apart from the human casualties, the destruction in- the examination of the synoptic setup during the event cluded approximately 3,000 houses partially or totally (Fig. 3). Regional- to local-scale conditions, related to

2138 | NOVEMBER 2019 AMERICAN METEOROLOGICAL SOCIETY NOVEMBER 2019 | 2139 Unauthenticated | Downloaded 10/05/21 05:16 PM UTC and 20–25 m s−1 at 850 hPa (not shown). At the surface, a 1,003-hPa low pressure system in the north at 1200 UTC (Fig. 3, white contours) was further deepening, enhancing the strong westerly flow over most of Greece. At this point it is worth mentioning that the typical high-wind regime over the Aegean Sea that also affects Attica during summer is the Etesians, which are north sector winds (Kotroni et al. 2001). Figure 4 shows the 10-min records of mean and maximum wind speed at the Penteli AWS. The station, denoted by the let- ter P in Fig. 1, is located at an altitude of 495 m and is 7 km Fig. 3. GDAS analysis of 500-hPa geopotential height (shading at 50-gpdm west of the ignition point. Two intervals), mean sea level pressure (white contours at 2-hPa intervals), distinct periods of high winds −1 and wind speed at 500 hPa (blue contours between 25 and 35 m s at are evident. The first period, −1 5 m s intervals) valid at 1200 UTC 23 Jul 2018. with mean wind speed exceeding 15–16 m s−1 and gusts reaching the development of downslope winds in the study area, 25 m s−1, was observed between 1230 and 1430 UTC, a were analyzed based on model simulations. 2-h window that included the time of the fire ignition The upper-level flow between 0000 and 1200 (1350 UTC). Then, between 1440 and 1520 UTC, the UTC 23 July 2018 was characterized by a positively wind speed decreased to ~10 m s−1, while the second tilted trough over the central Mediterranean, mov- 2-h period of high winds started right after, with mean ing eastward and interacting with an intensifying wind speed and gusts peaking at 18 and 26 m s−1, re- subtropical ridge downstream (Fig. 3). The net result spectively. During this last time window, the wildfire of this convergence was a subtropical jet stream ex- spread toward the sea. It should be noted that the tending from southern Italy toward Greece, with a wind gusts recorded during this day at many stations 25–35 m s−1 jet streak at 500 hPa (Fig. 3, blue contours) were the highest recorded since 2010 (when the AWS

Fig. 4. Temporal evolution of wind speed (mean and gusts) at Penteli AWS between 0600 and 2100 UTC 23 Jul 2018 (m s−1; at 10-min intervals). The time of the fire ignition is denoted with the flame while the duration of the fire event is shaded.

2138 | NOVEMBER 2019 AMERICAN METEOROLOGICAL SOCIETY NOVEMBER 2019 | 2139 Unauthenticated | Downloaded 10/05/21 05:16 PM UTC this part of the study, we only focus on the output of the atmosphere component of WRF-SFIRE at the highest resolution domain (i.e., 1 km). The simulated temperature over the central and eastern part of Attica, at 1400 UTC, ranged between 34° and 36°C, while it reached 38°–39°C in a narrow zone along the east- ern coast of Attica (Fig. 5). This is in agreement with observations from the Rafina and Nea Makri AWS (denoted by letters R and N in Fig. 1, respectively), where such high temperatures were recorded. The maximum daily temperature recorded at Rafina AWS during 23 July 2019 was 39°C (at 1230 UTC) and this was the highest tempera- ture recorded in Attica not only during that day, but also during all summer months of 2018. The narrow band along the eastern coasts of Attica where Fig. 5. WRF-SFIRE simulated fields at 1-km grid increment, valid at temperature exceeded 38°C was 1400 UTC 23 Jul 2018: 2-m temperature (color shading, following the also very dry with relative humid- color bar shown on the right), 2-m relative humidity below 20% (within ity (RH) less than 20% (Fig. 5). the white contour line), and 10-m sustained wind speed (wind barbs; Again, the model is consistent with one barb equals 5 m s−1, one half-barb equals 2.5 m s−1). The white observations, as the Rafina AWS line denotes the location of the cross section shown in Fig. 6, and the white box denotes the part of WRF-SFIRE inner domain that is shown recorded a minimum RH of 19% at in Figs. 8 and 9. 1230 UTC and RH remained below 30% for 10 consecutive hours (from network NOAAN started to expand in Attica) during 0950 to 1940 UTC). Finally, the model adequately any summer month (June–August). reproduced the wind field, showing sustained WNW To draw a more detailed picture of the weather winds around 15–16 m s−1 over the area of the wildfire situation during the event, the WRF-SFIRE modeling (denoted by the rectangle in Fig. 5). The model results system (Mandel et al. 2011) was employed. WRF-SFIRE shown in Fig. 5 clearly highlight the occurrence of the is a coupled fire-atmosphere modeling system that combination of high temperature, low relative humid- combines the Weather Research and Forecasting ity, and almost gale-force winds, conditions that set (WRF) numerical weather prediction model (Skama- the stage for the rapid spread of the wildfire. rock et al. 2008) with a two-dimensional fire spread A west–east cross section, following the thick line model (SFIRE; Mandel et al. 2009). For the purposes depicted in Fig. 5, is shown in Fig. 6. The cross section of this study, WRF-SFIRE was configured with three was drawn over the southern flanks of Mt. Penteli one-way nested modeling domains (25, 5, and 1 km), and crosses the ignition point as well as the Mati area telescopically nested and centered around the location to the east. At 1400 UTC 23 July, the wind vectors of fire ignition. The 0600 UTC forecast cycle of the within the cross section depict the downslope winds Global Forecasting System (GFS), available at the time toward the coast. Vertical velocity associated with this of ignition, was employed for initializing the modeling downslope flow was about −2.5 m −1s over the eastern system at 1200 UTC 23 July 2019, and the model was part of the slope. The relative humidity distribution allowed to integrate until 0600 UTC 24 July 2018. For across the cross section shows the drying of the

2140 | NOVEMBER 2019 AMERICAN METEOROLOGICAL SOCIETY NOVEMBER 2019 | 2141 Unauthenticated | Downloaded 10/05/21 05:16 PM UTC boundary layer, with RH less than 25% within the first kilometer of the troposphere and downstream of the mountain. Last, Fig. 7 presents the model- simulated continuous Haines in- dex (CHI; Mills and McCaw 2010) at 1400 UTC. CHI values provide a measure of atmospheric stabil- ity, indicating conditions that are favorable for the occurrence of extreme fire behavior. The highest CHI values, suggesting potential for uncontrollable fire behavior, were simulated to coincide spa- tially (Fig. 7) with the narrow band of high temperature and low humidity along the eastern coasts of Attica (Fig. 5).

FIRE SPREAD SIMULA-

Fig. 6. Vertical cross section valid at 1400 UTC 23 Jul 2018, following TIONS. Within the framework of the white line shown in Fig. 5, of relative humidity (color shading, WRF-SFIRE, fire spread is simu- following the color bar at the bottom), wind vectors within the cross lated with the semiempirical algo- section (scaled according to the vectors shown at the bottom right), rithm of Rothermel (1972), which −1 and vertical velocity (dashed lines; only the −2.0 and −2.5 m s contour employs the properties of the lines are shown). burning fuels along with terrain (slope) and meteorological data for computing the rate of spread (ROS). The coupling between the fire (SFIRE) and the atmosphere (WRF) occurs primarily through the sensible and latent heat fluxes released by the fire. For a detailed description of WRF-SFIRE, please refer to Mandel et al. (2011, 2014). WRF-SFIRE was configured with three one-way nested do- mains (25-, 5-, and 1-km grid increments). The fire spread simulation was carried out in a “fire” domain with grid spacing of 100 m, embedded within the highest-resolution “atmosphere” domain (i.e., 1 km). Fuels were represented with 15 standard fuel models derived from Scott and Burgan (2005) and additional cus- tom fuel models based on the de- Fig. 7. WRF-SFIRE simulated continuous Haines index at 1-km grid scription of the eastern Mediterra- increment, valid at 1400 UTC 23 Jul 2018. nean ecosystem (Dimitrakopoulos

2140 | NOVEMBER 2019 AMERICAN METEOROLOGICAL SOCIETY NOVEMBER 2019 | 2141 Unauthenticated | Downloaded 10/05/21 05:16 PM UTC Fig. 8. Fuel models used by the WRF-SFIRE model.

2002; Kalabokidis et al. 2013; Salis et al. 2016) that model. It should be noted that the use of the CORINE were selected following several test simulations. The land-cover dataset was of paramount importance for mapping of the selected fuel models was conducted us- the correct representation of the WUI on WRF-SFIRE ing high-resolution geospatial datasets of the Coper- simulation process. Test simulations without the use nicus Land Monitoring Service (CLMS; https://land of CORINE dataset failed to correctly advance the fire .copernicus.eu). Specifically, the datasets used include front within the WUI, which covers a large fraction the 100-m-resolution layers for forests (https://land of the domain. Figure 8 shows the final fuel models’ .copernicus.eu/pan-european/high-resolution-layers map for the study area. /forests) and grasslands (https://land.copernicus Figure 9 depicts the fire perimeter and midflame .eu/pan-european/high-resolution-layers/grassland), winds at 1500 and 2100 UTC from the WRF-SFIRE and Coordinate Information on the Environment simulation while the complete evolution of the simu- (CORINE) land cover (https://land.copernicus.eu lated wildfire is available in the online supplement. /pan-european/corine-land-cover). The forests One hour following the ignition of the fire (Fig. 9a), dataset was exploited for defining fuel models for the model simulation indicates a strong westerly flow the various forest types (i.e., conifer, broadleaf, and (velocities up to 8–10 m s−1) that was consistent with mixed forests), while the grasslands dataset was used concurrent observations (note that the presented for delineating fuel models related to Mediterranean wind field refers to the so-called midflame height). grasslands. Other fuel models (e.g., shrublands, More importantly, the simulated fire perimeter agricultural) were defined by remapping from the indicates that the model was able to reproduce the CORINE land-cover dataset. In particular, the sub- rapid fire spread. According to news reports and urban land-cover categories were combined with local evidence (Lekkas et al. 2018), the actual fire tree-cover density data for deriving the WUI fuel front reached Marathon Avenue (shown with the

2142 | NOVEMBER 2019 AMERICAN METEOROLOGICAL SOCIETY NOVEMBER 2019 | 2143 Unauthenticated | Downloaded 10/05/21 05:16 PM UTC north–south–oriented green line in Fig. 9) approximately 1 h following ignition that was timely reproduced by the model (Fig. 9a). By 2100 UTC (Fig. 9b) the simulated fire had already spread across most of its actual final perimeter. Further, the simulated total area burned (1,659 ha) was in good agreement with ob- servations (1,276 ha). The currently available computing infrastructure permits the timely execution of such rapid response fire spread forecasting systems. More specifically, the fire spread simulation of 6 h (24 h), with the setup of the current work, takes 15 min (1 h) clock time, using 200 cores of the computing infrastructure of the National Observatory of Athens. Thus, it is evident that WRF-SFIRE could be used for providing a fire spread forecast guidance.

CONCLUDING REMARKS. The deadly wildfire that affected Attica on 23 July 2018 was a typical example of how specific weather condi- tions can be conducive to extreme fire behavior. Analysis of the synoptic situation, together with examination of the available meteorological ob- servations, allowed for characterizing these con- ditions. Furthermore, application of the coupled fire–atmosphere WRF-SFIRE modeling system allowed for successfully simulating fire spread, following the proper representation of the fuels available in the area, with particular emphasis given on the WUI. The following conclusions should contribute to the better organization and preparedness of civil protection agencies in Greece, an effort which today is among the top priorities of the state.

• A dense network of AWS over the area (especially near WUI) is of paramount importance for the continuous monitoring of the prevailing near-surface conditions. Continuous measurements of temperature, relative humidity, and wind, through a dense network of surface stations, can pro- vide a real-time snapshot of the prevailing weather, which can be of great usefulness to fire managers before and during wildfires Fig. 9. Fire spread (red contour) simulated by WRF-SFIRE at (Clements et al. 2007). The AWS network (a) 1500 and (b) 2100 UTC 23 Jul 2018. The thick black line data can be also used for the calculation denotes the fire scar as defined by satellite data analysis. The green lines are used to delineate major roads in the and operational provision of fire weather study area. WRF-SFIRE 10-m wind speed at a 1-km grid indices (such an example is given in http:// increment is also shown. map.disarmfire.eu/Greece).

2142 | NOVEMBER 2019 AMERICAN METEOROLOGICAL SOCIETY NOVEMBER 2019 | 2143 Unauthenticated | Downloaded 10/05/21 05:16 PM UTC • A rapid response fire spread forecasting system, Dimitrakopoulos, A. P., 2002: Mediterranean fuel such as WRF-SFIRE, could be exploited for pro- models and fire behavior in Greece. Int. J. Wildland viding forecast guidance of high added value. Fire, 11, 127–130, https://doi.org/10.1071/WF02018. • Field measurements during large wildfires are Huang, C., Y. L. Lin, M. L. Kaplan, and J. J. Charney, necessary for a better understanding of the fire– 2009: Synoptic and mesoscale environments atmosphere interactions, following recent good conducive to forest fires during the October 2003 practices in the United States, such as described extreme fire event in Southern California.J. Appl. in Clements et al. (2018). Meteor. Climatol., 48, 553–579, https://doi.org • The synergetic use of the aforementioned obser- /10.1175/2008JAMC1818.1. vational and modeling tools helps to assess the Kalabokidis, K., P. Palaiologou, and M. Finnery, conditions that contribute to extreme fire behavior 2013: Fire behavior simulation in Mediterranean that often causes devastating impacts on the local forests using the minimum travel time algorithm. population and provides the necessary basis for the Proc. Fourth Fire Behavior and Fuels Confer- validation of coupled fire–atmosphere models. ence, Raleigh, NC, International Association of Wildland Fire, 468–492, www.iawfonline.org/wp In conclusion, the disastrous wildfire of Attica set -content/uploads/2018/02/4th_Fuels_Conference the stage for the overall redesign of strategies for fire _Proceedings_USA-Russia_updated_5.28.2015.pdf. danger warning and mitigation, a redesign which is Kotroni, V., K. Lagouvardos, and D. Lalas, 2001: The ef- of paramount importance in the context of increased fect of Crete island on the Etesian winds over the Ae- frequency of high fire danger conditions expected gean Sea. Quart. J. Roy. Meteor. Soc., 127, 1917–1938, under a changing climate. https://doi.org/10.1002/qj.49712757604. Lagouvardos, K., and Coauthors, 2017: The automatic ACKNOWLEDGMENTS. This research has been weather stations NOANN network of the National Ob- financed by the “Drought and Fire Observatory and servatory of Athens: Operation and database. Geosci. Early Warning System (DISARM) INTERREG BALKAN- Data J., 4, 4–16, https://doi.org/10.1002/gdj3.44. MEDITERRANEAN (2014–2020) project, which is Lekkas, E., and Coauthors, 2018: The July 2018 Attica cofunded by the European Union and national funds of (central Greece) wildfires—Scientific report (ver- the participating countries. The availability of the Global sion 1.0). Newsletter of Environmental, Disaster, and Forecasting System (GFS) and Global Data Assimilation Crisis Management Strategies, No. 8, National and System (GDAS) data is also greatly acknowledged. Kapodistrian University of Athens, Athens, Greece. Mandel, J., J. D. Beezley, J. L. Coen, and M. Kim, 2009: Data assimilation for wildland fires. IEEE Con- FOR FURTHER READING trol Syst. Mag., 29, 47–65, https://doi.org/10.1109 /MCS.2009.932224. Abatzoglou, J. T., R. Barbero, and N. J. Nauslar, 2013: —, —, and A. K. Kochanski, 2011: Coupled atmo- Diagnosing Santa Ana winds in Southern California sphere–wildland fire modeling with WRF 3.3 and with synoptic-scale analysis. Wea. Forecasting, 28, SFIRE 2011. Geosci. Model Dev., 4, 591–610, https:// 704–710, https://doi.org/10.1175/WAF-D-13-00002.1. doi.org/10.5194/gmd-4-591-2011. Clements, C. B., and Coauthors, 2007: Observing the —, and Coauthors, 2014: Recent advances and ap- dynamics of wildland grass fires: FireFlux—A field plications of WRF–SFIRE. Nat. Hazards Earth Syst. validation experiment. Bull. Amer. Meteor. Soc., 88, Sci., 14, 2829–2845, https://doi.org/10.5194/nhess 1369–1382, https://doi.org/10.1175/BAMS-88-9-1369. -14-2829-2014. —, N. P. Lareau, D. E. Kingsmill, C. L. Bowers, C. P. Mass, C. F., and D. Ovens, 2019: The Northern Califor- Camacho, R. Bagley, and B. Davis, 2018: The Rapid De- nia wildfires of 8–9 October 2017: The role of a major ployments to Wildfires Experiment (RaDFIRE): Obser- downslope wind event. Bull. Amer. Meteor. Soc., 100, vations from the Fire Zone. Bull. Amer. Meteor. Soc., 99, 235–256, https://doi.org/10.1175/BAMS-D-18-0037.1. 2539–2559, https://doi.org/10.1175/BAMS-D-17-0230.1. Mills, G. A., and L. McCaw, 2010: Atmospheric sta- Diakakis, M., G. Xathopoulos, and L. Gregos, 2016: bility environments and fire weather in Australia Analysis of forest fire fatalities in Greece: 1977–2013. – Extending the Haines index. CAWCR Tech. Rep. Int. J. Wildland Fire, 25, 797–809, https://doi.org/10 20, 158 pp., www.cawcr.gov.au/technical-reports .1071/WF15198. /CTR_020.pdf.

2144 | NOVEMBER 2019 AMERICAN METEOROLOGICAL SOCIETY NOVEMBER 2019 | 2145 Unauthenticated | Downloaded 10/05/21 05:16 PM UTC Nauslar, N., J. Abatzoglou, and P. Marsh, 2018: The Scott, J. H., and R. Burgan, 2005: Standard fire behavior 2017 North Bay and Southern California fires: fuel models: A comprehensive set for use with Ro- A case study. Fire, 1, 18, https://doi.org/10.3390 thermel’s surface fire spread model. USDA Forest /fire1010018. Service General Tech. Rep. RMRS-GTR-153, 72 pp., Rothermel, R. C., 1972: A Mathematical Model for www.fs.usda.gov/treesearch/pubs/9521. Predicting Fire Spread in Wildland Fires, USDA Skamarock, W. C., and Coauthors, 2008: A description Forest Service Research Paper INT-115, 40 pp., www of the Advanced Research WRF version 3. NCAR .treesearch.fs.fed.us/pubs/32533. Tech. Note NCAR/TN-475+STR, 113 pp., https://doi Ruffault, J., V. Moron, R. M. Trigo, and T. Curt, 2017: .org/10.5065/D68S4MVH. Daily synoptic conditions associated with large fire Werth, P. A., and Coauthors, 2016: Synthesis of knowledge occurrence in Mediterranean France: Evidence of extreme fire behavior: Volume 2 for fire behavior for a wind–driven fire regime.Int. J. Climatol., 37, specialists, researchers, and meteorologists. USDA 524–533, https://doi.org/10.1002/joc.4680. Forest Service General Tech. Rep. PNW-GTR-891, 258 Salis, M., and Coauthors, 2016: Predicting wildfire pp., www.fs.usda.gov/treesearch/pubs/50530. spread and behavior in Mediterranean landscapes. Whiteman, C. D., 2000: Mountain Meteorology: Funda- Int. J. Wildland Fire, 25, 1015–1032, https://doi mentals and Applications. Oxford University Press, .org/10.1071/WF15081. 355 pp.

The AMS Centennial weather history calendar is now available!

From over 600 submissions to the photo contest, our expert panel selected 14 photos depicting the amazing diversity of weather, water, and climate, and we have paired them with fascinating facts, records, and weather history events. Order now for yourself and for holiday gift giving! A portion of the proceeds will support the 100th Anniversary Campaign key area: Engaging the Public. Ametsoc.org/ 100Calendar

2144 | NOVEMBER 2019 AMERICAN METEOROLOGICAL SOCIETY NOVEMBER 2019 | 2145 Unauthenticated | Downloaded 10/05/21 05:16 PM UTC Communit� A member benefit from the AMS

Earn recognition for Network the contributions you make to the with colleagues using Community. Participate the comprehensive Member Directory. in focused and dedicated discussions relevant to you. Communit� AMS Community is our new online platform that will revolutionize the way you communicate, connect, and interact with your fellow Society members! We can’t wait to have you join the conversation! Activate your account with your existing AMS credentials today at: community.ametsoc.org

Questions? Please contact [email protected] | Downloaded 10/05/21 05:16 PM UTC